Abstract

The aviation industry plays a crucial role in the world’s transportation sector, and a lot of businesses rely on various airlines to connect them with other parts of the world. Flight delays are gradually increasing and bring more financial difficulties and customer dissatisfaction to airline companies. This research is done in three phases. In the first phase, Supervised Classification models like Logistic Regression, Decision Tree, Decision Jungle, and Neural Network Classifier are used to predict if the flight is delayed more than 15 minutes. Additionally, this phase highlights the key drivers – both controllable and uncontrollable – that globally contribute to the flight delay. In the second phase, Supervised Regression models like Linear Regression, Linear Regression with Principal Component Analysis (PCA) technique, and Neural Network Regressor model are used to predict the time by which the flight is getting delayed. The third phase of the research focuses on the trips that have major delays and identifies actionable, real-time insights into the key factors that contribute to the delays. This is achieved using the Explainable Artificial Intelligence (XAI) tool with the Local Interpretable Model-agnostic Explanations (LIME) package.

The Neural Network Classifier appears to be the best non-linear model for the dataset under consideration, with an accuracy of 92.5% and a precision of  73.6%. With the lowest RMSE of 3.52, MAE of 2.20, and maximum coefficient of determination(R square) of 93.44, the Linear Regression model with Principal Components stands out to be the best model for predicting delay in flight arrival time.

Conference Name: 3rd International Conference on Smart Technologies in Computing Electrical and Electronics (ICSTCEE 2022)